School District AI Policy Models: Ban, Scaffold, or Open?

In one season, three large systems answered the same question three different ways. New York City barred student-facing generative AI through eighth grade for roughly 600,000 kids. Katy ISD, outside Houston, banned chatbots through grade 6 and then opened supervised, cited access. Utah brought Gemini for Education to about 680,000 students statewide. These are the three school district AI policy models of 2026, and anyone who tells you the evidence favors one of them is ahead of the evidence. What the data does support is a claim about how kids use these tools, and each model handles that differently.

Key Takeaways

• Ban: New York City’s September 2, 2026 moratorium covers 2-K through grade 8 (about 600,000 students) for one year, with device-minute caps, companion chatbots barred district-wide, and five capped high school pilots.
• Scaffold: Katy ISD’s framework, effective 2026-27, bans open chatbots through grade 6, allows approved tools under supervision at grade 7, and permits generative AI in grades 8-12 with teacher authorization and required citation.
• Open: Utah’s State Board partnered with Google to bring Gemini for Education to roughly 680,000 students and 28,000 educators beginning 2026-27, at no cost, with district-by-district choice.
• Los Angeles is a fourth variant: a technical block on district devices for about 378,000 students, with policy recommendations due by the end of the 2026-27 year.
• No model has outcome data. PISA 2025’s finding that moderate, purposeful use tracks with better outcomes than either heavy or zero use is the closest thing to a standard any of them can be measured against.

The three school district AI policy models, described

A policy model here means the default access setting a district chooses for student-facing generative AI. Everything else, filters, training, integrity rules, follows from that choice.

Ban. New York City’s September 2, 2026 announcement set a one-year moratorium on student-facing generative AI for 2-K through grade 8, said it would discontinue or disable AI components in more than 38 previously allowed programs, and barred companion chatbots at every grade. Chalkbeat reported the device caps: no individual devices pre-K through grade 2, 30 minutes daily in grades 3-5, 45 in grades 6-8. High school gets five pilots (Quill, Edia, Brisk Teaching, Playlab, Intel AI-Ready Schools) with per-tool minute limits, a maximum of five classes per school, and two 45-minute AI critical-thinking modules a year.

Scaffold. Katy ISD, nearly 100,000 students, presented its framework July 20, 2026. Per Houston Public Media: K-6 may use district-approved, curriculum-embedded platforms like Amira Learning and Writable under teacher supervision but not open chatbots such as Microsoft Copilot or Google Gemini; grade 7 gains supervised access to approved tools; grades 8-12 may use generative AI for specific educational purposes with teacher authorization and must cite it. Assistant Superintendent Sanée Bell: “AI can support the learning process, but it cannot replace the thinking, the struggle, or the growth that learning requires.”

Open. Utah’s State Board of Education partnered with Google to deploy Gemini for Education across public schools starting 2026-27, reaching roughly 680,000 students and 28,000 educators at no cost. Interactions stay within a secure domain, data and conversations are not used to train models, and institutions retain administrative control. Crucially, adoption is local: officials described it as “LEA choice,” meaning districts and charters decide whether and how to use it.

Block, a fourth variant. Los Angeles Unified’s web filter began blocking generative AI on district devices for about 378,000 students in early September 2026, announced at the first meeting of an ad hoc committee whose recommendations are due by the end of the 2026-27 school year. A block without a policy is its own category, and it surprised board members.

Model to trade-off

Model	Example	What it does well	What it costs	What’s left to families
Ban	NYC, 2-K to grade 8, ~600,000 students	Removes unsupervised classroom exposure for young children; clear to enforce	Removes supervised practice too; high schoolers get 90 minutes of AI instruction a year	Nearly all AI judgment, at every age
Scaffold	Katy ISD, ~100,000 students	Matches access to developmental stage; citation makes AI use visible and gradeable	Depends heavily on teacher training and consistency across classrooms	Conceptual understanding before the grade 7 on-ramp
Open	Utah, ~680,000 students, LEA choice	Gives every student access under managed accounts; data not used for training	Access is not instruction; district-by-district variation is wide	Whether and how the tool is actually used well
Block	LAUSD, ~378,000 students	Fast; buys time while policy is written	Filter follows the district profile, so home and phone use is untouched	Everything the filter can’t reach, which is most of it
Consent	Florida rule, all 67 districts by July 2027	Parents get notice, choice, and a comparable non-AI alternative	Decision load shifts to families, tool by tool	Informed decisions about each tool
Mandate	Maryland, Idaho, Oklahoma, Virginia	Guarantees a written policy exists everywhere	A policy is not a practice; quality varies	Checking whether the policy reached the classroom

The rightmost column is the honest one. Every model leaves something for families, and the ban leaves the most.

What the evidence can and can’t settle

There is no outcome study comparing these models. What exists is a population-level finding about use patterns, and it cuts against both extremes.

PISA 2025, released September 8, 2026, found 46% of 15-year-olds across OECD countries use AI chatbots at least weekly for learning. Students who used AI daily to summarize assigned reading scored nearly 30 points below non-users in science, roughly a year of schooling. Students in the moderate band, about monthly to once or twice weekly, scored at or above non-users, and for the use “to help me learn,” weekly users scored highest. The OECD is explicit that these are associations that “may reflect a complex mix of who adopts AI and how they use it,” not causal proof. Secretary-General Mathias Cormann: technology “could strengthen learning only when used purposefully and not as a substitute for attention, effort, and understanding.”

Read plainly, the target is moderate and purposeful. A scaffold aims at that band by design. A ban pushes school exposure to zero while leaving personal-device use intact, and Common Sense Media’s “Teens in the AI Era” (August 18, 2026; n=1,017) found 59% of teens whose school blocked a tool simply switched to a personal device. An open model provides access without guaranteeing the pattern.

There is a genuine upside case too. A Harvard randomized controlled trial (Kestin et al., 2025, Scientific Reports) found students learned more from a well-designed AI tutor than from an active-learning class session. That is one tool, one course, carefully built. It says a good implementation can beat good teaching in a narrow case, not that access alone helps.

And the argument for caution with young children is real: the MIT Media Lab’s EEG study (Kosmyna et al., 2025) found LLM-assisted writers showed the weakest neural connectivity and lowest essay ownership. Small sample, preprint, adults not children. Suggestive, not settled.

The one number that every model has to answer for: only 27% of teens say a teacher has ever discussed safe AI use with them. Whatever your district chose, that is the gap.

What to do at home under each model

If your district banned it

Assume your child gets no supervised practice this year and plan accordingly. Fifteen minutes a week, verifying one AI-generated claim against a primary source, closes most of the instruction gap. Also confirm which classroom platforms lost features, because a ban on “student-facing generative AI” can quietly disable a reading or writing tool your child was using.

If your district scaffolded it

Know which rung your child is on and what changes next year. If they hit the grade 7 on-ramp next fall, teach the mechanism over the summer: a model predicts likely next text, it doesn’t look things up unless told to, and it states wrong answers as confidently as right ones. Start the citation habit two years before a rubric requires it.

If your district opened it

Access is the easy part; the pattern is the whole game. Set a household rule that matches the PISA finding: AI after a first attempt, never instead of one. Ask which teachers actually use the tool and for what, because “LEA choice” means your child’s experience depends on their building, not the state.

If your district blocked it without a policy

Ask two questions: which profile the block follows (in Los Angeles it follows the district account, not the hardware), and when the policy is due. Then treat the home rule as the operative one, because a filter that stops at the school Wi-Fi is not a parenting plan.

In every case, ask the same three questions

What AI tools are in use, who reviews any output that affects my child’s grade or placement, and what will my child be taught about AI this year and in which class. Those three answers describe your child’s actual situation better than the model label does. Our one-page home AI policy template covers the household side.

What not to do

Don’t pick a side based on the label. A well-run ban with strong home support beats a sloppy open rollout, and a thoughtful scaffold beats both if the teacher training is real. And don’t treat any of these as settled science; the districts themselves don’t. Katy ISD calls its framework a “living resource,” New York City’s moratorium runs one year pending a coalition’s recommendations, and Los Angeles has no policy yet at all.

What to Watch For Over the Next 3 Months

• Week 4: Identify which model your district actually uses, in writing, and which platforms changed this fall. The label in a press release and the experience in a classroom are often different.
• Month 2 red flags: A high schooler in a pilot who can’t name the tool or its limits. A scaffolded district where two teachers in the same building give contradictory rules; Common Sense Media found 28% of teens report exactly that. An open rollout with no teacher training attached.
• Month 3 self-check: Can your child describe their school’s rule in one sentence and name one thing AI reliably gets wrong? Watch for the first outcome reports: LAUSD’s committee recommendations are due by the end of the 2026-27 year, and New York City’s moratorium expires with a coalition report.

Frequently Asked Questions

Which of these school district AI policy models works best?

Nobody knows yet, and any confident answer is ahead of the evidence. PISA 2025’s association between moderate, purposeful use and better outcomes is the closest available standard, and a scaffold aims at that band most directly, but no district has published outcome data comparing approaches.

Does a ban actually stop kids from using AI?

It stops school-supervised use. Common Sense Media found that among teens whose school blocked a tool, 59% switched to a personal device. A ban shapes where use happens more than whether it happens.

Is Utah’s open model risky for student data?

The stated terms are better than most: interactions stay in a secure domain, data and conversations are not used to train models, and institutions keep administrative control. The open question is instructional, not privacy: access without training doesn’t produce the use pattern the data favors.

Why does Katy ISD’s citation requirement matter?

Because it makes invisible use visible. Common Sense Media found that among teens who get answers from AI, 31% rewrite them to sound like their own. A required citation converts that into a disclosed, gradeable act a teacher can respond to.

My district hasn’t decided. What should I ask for?

A written policy, a list of AI-enabled tools, a limit on AI in grading, discipline, placement, and special-education decisions, a real opt-out with a comparable alternative, and a grade-level plan for what students will learn. Those five items appear across the strongest 2026 policies. We mapped the legal landscape in 134 AI education bills in 31 states.

How long until we know which model is better?

Longer than a school year. New York City’s moratorium runs through 2026-27 with a coalition report to follow; LAUSD’s recommendations are due by the end of that year; Katy ISD calls its framework a living resource. Meaningful outcome comparisons would need multiple years and better measures than test scores alone.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Office of the Mayor of New York City. (2026, September 2). “Mayor Mamdani and Chancellor Samuels Put Students First with Nation’s Broadest Generative AI Moratorium in Schools.” https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat
2. Chalkbeat New York. (2026, September 2). “NYC to ban student AI tools in 2-K through 8th grade and limit classroom screen time.” https://www.chalkbeat.org/newyork/2026/09/02/nyc-schools-to-set-ai-policy-ban-screen-time-limits/
3. Houston Public Media. (2026, July 24). “Katy ISD restricts the use of AI in elementary school classrooms.” https://www.houstonpublicmedia.org/articles/education/2026/07/24/557895/katy-isd-ai-restrictions-elementary-school/
4. Utah State Board of Education. (2026, June). “USBE adopts Gemini for Education to enhance student learning and teacher support.” https://www.schools.utah.gov/publicrelations/_publicrelations_/_2026/USBEAdoptsGeminiEducationPressRelease.pdf
5. LAist. (2026, September). “LAUSD students barred from AI use, in a surprise to school board and parents.” https://laist.com/news/education/lausd-students-barred-artificial-intelligence-tools
6. OECD. (2026, September 8). PISA 2025 Results (Volume I). https://www.oecd.org/en/publications/pisa-2025-results-volume-i_73451bc5-en.html ; AI-use findings at https://technode.global/2026/09/08/students-avoiding-ai-for-schoolwork-outscore-peers-in-science-bar-one-use-oecd/
7. Common Sense Media. (2026, August 18). Teens in the AI Era: Schoolwork and the Skills That Matter. https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter
8. Kestin, G., et al. (2025). “AI tutoring outperforms in-class active learning.” Scientific Reports. https://www.nature.com/articles/s41598-025-97652-6
9. Kosmyna, N., et al. (2025). “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv 2506.08872. https://arxiv.org/abs/2506.08872
10. Central Florida Public Media. (2026, September 16). “Florida approves statewide AI rules for K-12 schools and colleges.” https://www.cfpublic.org/education/2026-09-16/florida-approves-statewide-ai-rules-for-k-12-schools-and-colleges
11. K-12 Dive. (2026, July 9). “4 more states require districts to adopt AI policies.” https://www.k12dive.com/news/4-more-states-require-districts-to-adopt-ai-policies/824749/